基于双近邻标准化和PCA的多阶段过程故障检测

现代工业产品的生产往往需要多个生产阶段，多阶段生产过程的故障检测成为一个重要问题。多阶段过程数据具有多中心、各工序数据结构不同等特征。针对多阶段过程数据的特征，提出了基于双近邻标准化和主元分析的故障检测方法（DLNS-PCA）。首先寻找样本的双层局部近邻集；其次使用双层局部近邻集的信息标准化样本，得到标准样本；最后在标准样本集上使用主元分析方法进行故障检测。双局部近邻标准化能够将各阶段数据的中心平移到同一点，并且调整各阶段数据的离散程度，使之近似相等，从而将多阶段过程数据融合为服从单一多元高斯分布的数据。进行了青霉素发酵过程故障检测实验，实验结果表明DLNS-PCA方法相对于PCA、KPCA、FDkNN等方法对多阶段过程故障具有更高的检测率。DLNS-PCA方法提高了多阶段过程故障检测能力。

DLNS-PCA-based fault detection for multimode batch process

Modern industrial products often require multiple production stages, and the fault detection of multi-stage production process has become an important issue. Multi-stage process data have the characteristics of multi center, different structure of each stage and so on. Aiming at the characteristics, a fault detection method based on double local neighborhood standardization and principal component analysis (DLNS-PCA) is proposed. Firstly, the double local neighborhood set of the sample is found. Secondly, the standard samples are obtained by using the information of the double local neighborhood set. Finally, the PCA method is used to detect the fault on the standard sample set. The DLNS can move the data centers of each stage to the same point, and adjust the degrees of dispersion of data at each stage to make its approximately equal, then multi-stage process data is fused to a single modal data following multivariate Gauss distribution. A fault detection of penicillin simulation process was carried out. The results showed that DLNS-PCA has higher fault detection rate than PCA, KPCA and FD-kNN methods. DLNS-PCA method improves the efficiency of multi-stage process fault detection.